Perplexity Orchestrator API launches with 45 models and one bill
Perplexity opened its model layer as a paid API, with no provider keys and open-weight models it serves itself.
Perplexity launched the Orchestrator API on November 17, 2026. It gives developers one API key and one bill for models from Anthropic, OpenAI, Google, xAI, Z.ai, DeepSeek and Moonshot AI, and retries failed requests on another deployment of the same model. Perplexity adds no platform fee on top of its API credits.
What changed
- The base URL is
https://api.perplexity.ai/orchestrator/v1and requests use an existing Perplexity API key. Every catalog model accepts the OpenAI Chat Completions, OpenAI Responses and Anthropic Messages formats, with prefixed IDs such asopenai/gpt-6-lunaorperplexity/glm-5.3. - The models page lists 45 models on November 18, 2026. Open-weight models (GLM 5.3, GLM 5.3 Flash, DeepSeek V4 Pro 0813, Kimi K3) are served by Perplexity or its inference partners. GPT-6.2 Astra and Sol are not listed yet.
- Billing is per token at each model's published rate, drawn from Perplexity API credits, with no per-request fee. The API does not accept OpenAI, Anthropic or other provider keys.
- On a provider error, rate limit or timeout, the request is retried on another deployment of the same model. Requests that fail before any output are not billed; a stream that fails after output starts is billed for the tokens delivered.
- Google, xAI and open-weight models are open from usage Tier 0, OpenAI and Anthropic models from Tier 1, and Claude Opus 5, Opus 5.5 and Fable from Tier 2. Organizations with zero data retention cannot call the Claude Fable models.
- The API is stateless:
store: true,previous_response_idandconversationon/responsesreturn HTTP 400. Routes (perplexity/auto, billed at the rate of the model that answered) and Auto effort are in preview.
Orchestrator prices
Open-weight models, per million tokens from Perplexity's Orchestrator models page on November 18, 2026:
| Model | Input | Output | Cache read |
|---|---|---|---|
perplexity/glm-5.3 | $1.40 | $4.40 | $0.26 |
perplexity/glm-5.3-flash | $0.15 | $0.50 | $0.03 |
perplexity/deepseek-v4-pro-0813 | $1.32 | $3.96 | $0.044 |
perplexity/kimi-k3 | $3.00 | $15.00 | $0.30 |
Selected frontier models from the same page: openai/gpt-6.1-sol $2.00/$10.00 (up to 272K input), anthropic/claude-sonnet-5-5 $2.00/$10.00, and google/gemini-3.8-flash $0.75/$3.75, a rate Perplexity applies through December 31, 2026.
On RouterPlus
The same four open-weight models in the RouterPlus catalog, per million tokens on November 18, 2026. Lowest input and lowest output are separate minima and can come from different providers.
| Model | Providers | Lowest input | Lowest output |
|---|---|---|---|
| GLM 5.3 | 15 | $0.20 | $1.80 |
| GLM 5.3 Flash | 14 | $0.05 | $0.20 |
| DeepSeek V4 Pro 0813 | 7 | $0.92 | $2.08 |
| Kimi K3 | 11 | $0.60 | $9.50 |
RouterPlus sends each request to a provider of the model you pick and fails over between providers of that model. Calls with your own OpenAI, Anthropic, Azure or AWS Bedrock key carry no RouterPlus fee (bring your own key); other calls use prepaid credits at listed model rates. The API base is https://api.routerplus.com/v1.
Sources
- Perplexity: Introducing Orchestrator: One API for frontier and open-weight models (November 17, 2026)
- Perplexity API docs: Orchestrator API quickstart
- Perplexity API docs: Orchestrator models & pricing
- Perplexity API docs: Orchestrator API FAQ
- Perplexity API docs: Routing & reliability
- RouterPlus model catalog (November 18, 2026)
